Copperhead brings agentic AI to PCB design with gated, commit-per-stage pipeline

Copperhead is an open-core CLI that applies the agentic coding pattern to hardware design, editing KiCad files directly rather than generating code. From a plain-language brief it runs eight sequential stages—spec, architecture, parts selection, schematic, layout, fabrication outputs, firmware, and a bring-up plan—each executed as its own agent run. Every stage is gated: it must write a real artifact to disk (a filled-in budget, a datasheet-justified BOM, ERC- and DRC-clean design files verified by KiCad itself) and land as a single commit before the next stage begins. If a gate fails, the run halts.

The pitch targets ‘drift,’ the way a single design decision fragments across a schematic, BOM, power budget, and documentation until an inconsistency surfaces at bring-up—where a respin can cost $5,000 to $50,000 and burn six to eight weeks. Copperhead’s guardrails are notably conservative for an AI tool: it refuses to run on a dirty git tree, won’t touch a design file before a validated change proposal exists, blocks edits that violate a documented budget or constraint (citing the offending line), and won’t invent an unjustifiable part number. Edits are surgical changes to KiCad’s s-expression source so diffs stay small and untouched parts of the file remain byte-identical, keeping changes reviewable.

The business model is open-core: the CLI is free, runs locally, uses your own Claude or GPT-5 API key, and produces plain markdown, JSON, and KiCad output to avoid lock-in. Paid tiers start at $49 per user/month for hosted cloud runs with a web viewer, scale up to team governance with a CI bot that gates pull requests, and top out at self-hosted or VPC enterprise deployments with Altium support, RBAC, and an audit trail. The whole framing—gates, commits, clean checks, and a compliance-ready trail—positions AI hardware design as something an auditor can actually verify.